Concept hierarchies and Credibility

Transcription

1 Jean Mark Gawron Alex Dodge Kathleen Burke August 7, 2013

2 1 Introduction Group studies Ontologies and concept hierarchies 2 Groups and conceptual frameworks 3 Credibility and importance 4 Tools

3 Group studies Studying groups organized for collective action Project task Task 3: Acquire the vocabulary and concept hierarchies for groups organized around ideas or themes or for collective action. 1 What are the defining features of the group? Do they constitute a real world community? An online community? 2 What are their goals? 3 What are their ideas? 4 What are they likely to do? 5 What are the characteristic features of their language and their discourse? What doo they talk about? How do they talk about it?

4 Group studies Three groups (kinds of groups) Website data collected from Google queries, mixture of blogs, forums, and organization websites 1 White militants 2 Vaccine debates [anti vaccine movement, pro vaccine websites, mostly professional or government] 3 Climate change [pro and anti]

5 Group studies Two applications 1 Word clustering 2 Relation extraction with Random Fields

6 Ontologies and concept hierarchies Ontologies Wikipedia definition... represents knowledge as a set of concepts within a domain, and the relationships between pairs of concepts. It can be used to model a domain and support reasoning about concepts.... a formal specification of a shared conceptualisation... provides a shared vocabulary, which can be used to model... the type of objects and/or concepts that exist, and their properties and relations. consists-of(water, [oxygen, hydrogen]) subtype( car-make, [Volvo, Volkswagen, BMW])

7 Ontologies and concept hierarchies Relation extraction consist-of 1 Water consists of hydrogen and oxygen. 2 Water is made up of hydrogen and oxygen. 3 Hydrogen and oxygen combine to make water. 4 Hydrogen bonds with oxygen to make water. 5 Hydrogen combined with oxygen makes water. Group discourse analogues: Ideology components What are the salient relations in the data? What are the texts talking about?

8 The vocabulary-based method 1 Collect data on group documents (websites, blogs, Tweets) 2 Characterize each word usage patterns in a set of documents, using syntactic analysis 3 Clustering words with with similar distributions together 4 A hierarchy of words suggesting concepts and conceptual connections

9 White Militant Data I An example Blood & Honour is a world-wide pan Aryan organization dedicated to the struggle for survival and prosperity of the White Race. Since the Martyr s death of its founder, Ian Stuart Donaldson, the Movement has experienced a period of turmoil and turbulence. Profiteers and pop stars have -unsuccessfully - tried to abuse B&H for their own selfish puposes. ZOG has geared up the campaign of oppression and the media has constantly smeared the national revolutionary forces. Members and supporters all over the White world have often been misinformed through infiltrators or left on their own due to the problems of the main organization...

10 White Militant Data II experienced_17 prep_since nsubj aux dobj death_5 Movement_15 has_16 period_19 poss prep_of det det prep_of Martyr_3 founder_8 the_14 a_18 turmoil_21 prep_of det poss appos conj_and the_2 its_7 Donaldson_12 turbulence_23 nn nn Ian_10 Stuart_11 Since the Martyr s death of its founder, Ian Stuart Donaldson, the Movement has experienced a period of turmoil and turbulence.

11 White militant data III identity folk nationalism democracy existence front freedom future right center genocide murder fraud assassination crime act violence abuse activity behavior propaganda agenda racism immigration threat terror action force establishment enemy war attack campaign struggle death crisis disaster return destruction change revolution body gun plane ground hand head spirit field seed fear desire love heart end beginning cross light blood

12 Climate change judgments Target/Importance Global climate change, conversely, may not have such a substantially negative effect on the world s population, standards of living, health, and survival. Target/Sentiment When you start to explore the issue, climate change can be terrifying. Information/Source Million Climate Jobs Campaign has released a new booklet describing how the transformation of the economy to one that protects people and the environment, can confront two of South Africas biggest threats: climate change and unemployment Target/credibility What is true is that climate change is real and its already happening.

13 Credibility judgments 1 X is credible 1 informational targets: pronouncements, claims, stories 2 sources 2 Kinds of judgments Debunking Calling into question the credibility or reliability of a story Legimitizing Affirming the credibility of a source

14 Tasks Wiebe (2003) [sources, predicates, MPQA corpus] Sources 1 Identify Targets Predicates 2 Infer polarities (positive/negative credibility)

15 Key sentences: Climate change Anyone who says they can confidently predict global climate change or effects is either a fool or a fraud. Tags Definition Source Source of information (thought, speech, writing) Predicate Word or phrase expressing credibility, importance, or sentiment opinion (positive, negative) Target The message or phenonemon about which the opinion is held, by the source, if there is a source, or by the author; simultaneous judgments possible: lies emanating from the White House

16 Example I [ source Anyone who says [ source they] can [ pred opinion(credibility,positive) confidently] [ pred opinion(credibility,positive) predict] [ target global climate changes or effects] ] is either a [ pred opinion(credibility,negative) fool] or a [ pred opinion(credibility, negative) fraud].

17 Example II [ target That is how [ source they] [ pred (credibility,positive,negative) make it seem] [ target that all the top scientists are agreed,] ] [ source he] [ pred (credibility,positive) says].

18 Some annotated data

19 Predication by predication

20 Factoring the problem He said she said global warming is a myth. Nested brackets: [ source He] [ pred said] [ target [ source she] [ pred said] [ target [ target global warming] is a [ predicate myth].]] Shallow brackets: [ source He] [ pred said][ source she] [ pred said] [ target global warming] is a [ predicate myth]. Tags: He/B-Source said/b-pred she/b-source said/b-pred global/b-target warming/i-target is/o a/o myth/b-pred./o

21 Conditional Random Fields